CE-TransUnet

This Repository is Code for 'CE-TransUnet: A Convolutional Enhanced Model for Pulmonary Alveolus Pathology Image Segmentation' on ICIC. Link: http://poster-openaccess.com/files/icic2024/1949.pdf

You could follow the instruction below to reproduce our project:

Dataset Configuration

Under the data folder, the directory format is as follows:

• The JPEGImages folder stores original images.
• The mask_input folder stores mask images. Original and mask images share the same filenames.

Preprocessing

Run mask_input_trans.py to convert pixel value 255 in mask images to 1.

Configuration Adjustment

Open utils.py and modify the input image size, preferably in multiples of 224x224.

Open ce_net.py and adjust the following values if needed:

Training

Open train.py and modify the following values:

CE_TransUnet and CE_TransTest are optional.

Simply run train.py to initiate training.

Testing

Open test.py and make necessary modifications.

Import the weight files into the params folder.

Run test.py to execute testing.

Post-training Transformation

Run res_trans.py.